Documentation

Appearance

VSLLabs ​

February 17, 2025

The "Specks of Light" podcast episode 161 features Tal Meged. Since the podcast is in Hebrew, we transcribed the episode, and asked ChatGPT to summarize it.

Introduction to Tal Meged & VSL ​

The podcast opens with an introduction to Tal Meged, a key figure behind VSL (Visual Sign Language), a company dedicated to revolutionizing accessibility through AI-driven sign language translation. Meged shares his personal and professional journey, which began in the Israeli Defense Forces in Unit 81, where he worked on advanced AI systems. This military experience laid the groundwork for his later ventures in business intelligence (BI) and data storage.

Tal's journey also led him to VSL, a company that uses deep learning and real-time 3D animation to translate text into sign language. VSL's mission is to bridge the communication gap for the deaf and hard-of-hearing communities.

The Problem & VSL’s Mission ​

The Challenge:
One of the core problems that VSL addresses is the lack of accessibility for people who are deaf or hard of hearing. Sign language, unlike spoken languages, is often underserved by digital platforms. This leads to difficulties in communication in many sectors like public transportation, healthcare, and customer service.

VSL's mission is to make real-time sign language communication accessible through AI avatars, making it possible for users to seamlessly interact with technology in their native sign language. They aim to create hyper-realistic avatars that convey not just signs but also context, emotion, and intent in the most natural way possible.

VSL’s Technological Approach ​

VSL’s technology can be broken down into three core components:

1. Natural Language Understanding (NLP & Glossing) ​

The first step in VSL’s process is to translate written or spoken text into structured sign language. This is achieved using natural language processing (NLP), where the system interprets the meaning of the text and glosses it—i.e., converts it into a format that reflects the grammar of sign language.

Key points discussed:

  • Sign languages are not direct word-for-word translations of spoken languages, meaning that VSL's system needs to understand context.
  • Ambiguities and slang are handled by VSL’s models, making sure that meanings, like "rock" (music vs. stone), are appropriately differentiated.

2. Pose Estimation & Computer Vision ​

Once the text is translated into a form that makes sense in sign language, the next step is to create the visual avatar that will deliver the translation. This is done using pose estimation models, such as MediaPipe, which track human body movement to generate gestures.

  • Key consideration: VSL uses custom-built datasets to ensure the signs are as natural and accurate as possible, incorporating facial expressions and hand positioning that are essential for conveying accurate meaning in sign language.

3. 3D Animation & Avatar Rendering ​

The final step is to render the signs using 3D animation. VSL uses Unity, a game engine, to create real-time animated avatars. These avatars are designed to be lifelike, moving fluidly and mimicking human body language rather than looking stiff or robotic.

  • The use of Unity allows VSL to provide realistic avatar-based translations on various platforms.

This system results in hyper-realistic, fluent sign language avatars that can perform real-time translations of text into sign language, delivering accurate communication.

Target Customers & Market ​

VSL primarily operates in a B2B (business-to-business) model, targeting enterprises that need accessible communication solutions. The podcast highlights some of VSL’s primary customer sectors:

1. Transportation (Airlines & Airports) ​

One of VSL's key markets is the transportation sector. Through its VSL Aviation Sense product, the company provides real-time sign language translations for things like flight updates, announcements, and safety instructions. This helps ensure deaf or hard-of-hearing passengers can navigate transportation hubs like airports without difficulty.

2. Healthcare & Customer Service ​

Another critical market for VSL is healthcare. Hospitals, insurance companies, and medical service providers use VSL to enhance accessibility, allowing deaf patients to interact with digital platforms and customer service representatives. For example, chatbots powered by VSL’s technology can help users schedule appointments or understand insurance policies, all via sign language.

3. Public Sector ​

Government and public sector organizations also form a significant portion of VSL’s customer base. In public spaces like train stations and airports, VSL’s avatars are used in digital kiosks, helping visitors interact with signage and emergency alerts in sign language.

4. Entertainment & Education (Future Expansion) ​

In the future, VSL is looking to expand into entertainment and education. This could include providing real-time sign language translation for platforms like Netflix, YouTube, and educational lectures.

Technological Pipeline & Future Developments ​

While VSL’s current technology stack is impressive, there are plans for further improvement and expansion:

  • Real-Time Streaming: VSL is working on improving streaming capabilities for live events such as conferences, meetings, and broadcasts, making sign language accessible for real-time viewers.

  • Multilingual Expansion: Currently focused on American Sign Language (ASL), VSL plans to expand to other sign languages such as British Sign Language (BSL), German Sign Language, and Brazilian Sign Language.

  • Enterprise Integration: VSL aims to integrate its technology with major platforms and applications, such as enterprise software and web accessibility standards (WCAG AAA), to ensure wider accessibility.

Challenges in the Industry ​

Despite VSL’s success, several challenges exist that hinder faster progress:

  • Data Scarcity: One of the most pressing challenges is the lack of large-scale datasets for training AI models in sign language. This scarcity means VSL has to build its own datasets and continuously improve them.

  • Variety of Sign Languages: There is no universal sign language, and over 200 different sign languages exist across the globe. VSL’s system must therefore be adaptable to different linguistic systems, adding complexity to their solution.

  • Regulatory Hurdles: While captioning is often required, sign language translation remains largely unregulated, especially in the private sector. Without clear mandates, widespread adoption of accessible technologies is slower.

  • Enterprise Sales: Selling to large enterprises can be challenging, as decision-makers often need more than just corporate social responsibility arguments—they need to see clear business value.